What resolution does CodeProject.AI pass from Blue Iris to YOLOv5?

[MarineSands]

Hi all,

I'm working on custom training with YOLO v5 6.2 and want to be sure the images in my training set are the optimal resolution. I've seen the settings in Blue Iris regarding AI, but I haven't seen any reference to the actual resolution that is passed to the model during inference. As a result, I don't know if the frames are the same as in BI or if they're resized somewhere along the way before actual inference. Any guidance on this, or pointers to documentation where I can learn, would be really appreciated.

[MikeLud]

Hi all,

I'm working on custom training with YOLO v5 6.2 and want to be sure the images in my training set are the optimal resolution. I've seen the settings in Blue Iris regarding AI, but I haven't seen any reference to the actual resolution that is passed to the model during inference. As a result, I don't know if the frames are the same as in BI or if they're resized somewhere along the way before actual inference. Any guidance on this, or pointers to documentation where I can learn, would be really appreciated.

The all the YOLO v5 are trained with image at 640x640 resolution. If you send an image with greater resolution the image gets resized to 640x640 before actual inference. The below link has details on training. You can increase the resolution when training using the img parameter

https://docs.ultralytics.com/yolov5/tutorials/train_custom_data/

[MarineSands]

Thank you!

[timdavid010]

YOLOv5 (v6.2) automatically resizes images for inference, typically to 640x640 (default). If you're using Blue Iris (BI), check resizing frame first

you can check through this process

1. BI Settings – Ensure BI's AI settings match your desired resolution.
2. YOLO Input Size – Check the imgsz parameter in training/inference (--img 640 or custom).
3. Test Inference – Run a test and print image shape after processing:
   python "print(results.ims[0].shape) # Check YOLO-processed image size "
   cross check then
   1)Match BI settings with YOLO training resolution.
4. Train at a higher imgsz (e.g., 1280) for accuracy or lower for speed.
5. If unsure, BI forums might help with its internal processing.